Gemma-4-31B-it-assistant and Gemma-4-31B-IT-NVFP4 models now available on Amazon SageMaker JumpStart
Amazon SageMaker JumpStart now includes Google DeepMind's Gemma-4-31B-it-assistant and NVIDIA's Gemma-4-31B-IT-NVFP4 models, expanding the portfolio of foundation models for AWS customers. These models offer both full-precision and optimized quantized variants of the flagship 31B dense architecture for high-performance, scalable AI solutions on AWS infrastructure.
Google DeepMind's Gemma-4-31B-it-assistant and NVIDIA's Gemma-4-31B-IT-NVFP4 models are now available on Amazon SageMaker JumpStart. These models adapt Google's flagship 31B dense architecture for enterprise workloads, offering both full-precision and optimized quantized variants. The Gemma-4-31B-it-assistant specializes in multimodal inference, coding, and agent workflows, accepting text and image inputs (video as frame sequences) and outputting text, supporting a 256K token context window and over 140 languages, ranking in the top three open models on the Arena AI text reader board. The Gemma-4-31B-IT-NVFP4 model is quantized to 4-bit FP4 precision using NVIDIA's ModelOpt framework, reducing memory usage to about 18.5 GB (68% smaller than the base model) and speeding up inference by approximately 2.5x while maintaining 97-99% of the original model's quality. These models are ideal for cost-effective, high-throughput production deployments on NVIDIA RTX, DGX Spark, and data center GPUs. Using SageMaker JumpStart, customers can deploy these models in a few clicks to address specific AI use cases. Access the SageMaker JumpStart model catalog via the SageMaker console or deploy models to your AWS account using the SageMaker Python SDK. For details on deploying and using foundation models with SageMaker JumpStart, see the Amazon SageMaker JumpStart documentation.
Why it matters
Amazon SageMaker JumpStart is a platform that allows AWS customers to access AI foundation models, and the addition of these new models makes it easier to deploy enterprise AI workloads.